This is an outdated version published on 2021-10-18. Read the most recent version.
Preprint / Version 1

Study of appropriate offload destination based on software arithmetic intensity

##article.authors##

  • Yoji Yamato

DOI:

https://doi.org/10.31224/osf.io/jmr62

Keywords:

Arithmetic Intensity, Automatic Offloading, Environment-Adaptive Software, Evolutionary Computation

Abstract

IEICE Technical Workshop on Network Software, BS-26-2. I have proposed environment-adaptive software that automatically converts the description code according to the destination device and operates it with high performance. Currently, for environments where GPUs, FPGAs, and manycore CPUs are mixed in the cloud or so on, all offload destination performance is verified and the fastest migration destination is selected, and it takes a long time. In this paper, I use the Arithmetic Intensity of the application loop statement to verify whether it is possible to select for GPU or manycore CPU.

Downloads

Download data is not yet available.

Downloads

Posted

2021-10-18

Versions